What We Do: Signal Processing with Machine Learning

When we launched Expedition Technology (EXP) in 2013, our vision of the future was not centered on a specific technology. Building a company that is aimed to grow for decades involves more than just placing a bet on the success of a single specific technology. Rather, our goal was to construct an organizational culture that empowers talented teams of people to solve our customers’ hardest problems. For this reason, most of the blog posts we’ve shared to date have focused on our culture. This article is the first in an ongoing series entitled, “What We Do.” The series will explore the work we perform at EXP and how we envision things evolving over the coming years.

The first EXP employees had strong digital signal processing (DSP) expertise, so it is no surprise many of our initial successes and ongoing growth are in that technical area. The deep learning revolution started right around the same time as EXP, so the potential to explore novel solutions to traditional signal processing with machine learning technology was and continues to be compelling. We have been at the forefront of developing and deploying these new approaches to signal processing, most notably in two large R&D programs that led to work in many other ongoing EXP programs:

  • DARPA’s Radio Frequency Machine Learning Systems (RFMLS: 1, 2) program beginning in 2018, whose goal was to develop the foundations for applying data-driven machine learning to the RF spectrum domain, especially in the areas of spectrum understanding, emitter fingerprinting, and radio control.
  • IARPA’s Securing Compartmentalized Information with Smart Radio Systems (SCISRS) program, which aims to develop smart radio techniques that can automatically detect and characterize anomalous RF signals.

EXP was the only performer on RFMLS selected for all phases of the program in both technical areas, and we were recently notified we are one of three performers selected to continue to the second phase of SCISRS. These successes are propelled by the skills of the engineers we hire at EXP, their creativity developing cutting-edge machine learning solutions that leverage deep domain expertise in DSP to outperform traditional hand-tuned approaches, and our culture of continuous learning, agile development, teamwork, and support that provides the opportunity for those engineers to do their best work.

As we continue to pursue and define the future of machine learning for digital signal processing, our key areas of current and future growth span a broad set of algorithms, system building, and collaboration for our customers.

RF Fingerprinting: While we developed state-of-the-art deep learning algorithms for device-level fingerprinting on RFMLS (see our publications page for details), our solutions for uniquely identifying emitters continue to improve and address more sophisticated scenarios, thereby remaining an important aspect of our ongoing research and development. Deployed solutions of this fingerprinting technology address key Department of Defense (DoD) and Intelligence Community (IC) priorities and make communication systems more secure and resilient.

RF Scene Understanding, and Novel Signal Detection and Classification: In a complex RF environment, signals of interest mix with background or less relevant signals and are corrupted by channel and noise. Traditional algorithms for finding signals of interest are typically flummoxed in a complex signal environment where the definition of “novel or interesting” cannot be mathematically described. Our team continues working on deep learning and statistical techniques that effectively separate and classify signals, even in low signal-to-noise ratio (SNR) environments, to find novel signals of interest. Our progress in this area will be invaluable in applications such as wireless communication, surveillance, and electronic warfare.

Deploying Models on Novel Platforms: DSP solutions are often deployed in low size, weight, and power “edge” environments and need to run near real time in potentially very high data rate sensors to be reactive to swiftly changing environments. As a result, EXP actively develops a variety of solutions for running deep learning algorithms on heterogeneous, embedded processing architectures with varying throughput, power, and bit depth capabilities.

Ongoing Collaboration with Academia and Industry Partners: At EXP, we believe collaboration is key to driving innovation in the field of digital signal processing. We maintain strong partnerships with academic institutions and other industry leaders to share knowledge, ideas, and best practices. This collaborative approach is essential for remaining at the forefront of technological advancements and knowing how to apply them to deliver cutting-edge solutions to our customers to solve their most complex challenges.

EXP remains committed to pushing the state of the art in digital signal processing. Our talented engineers and researchers, supported by our culture of continuous learning, are dedicated to developing innovative solutions that address the challenges faced by our customers.

We are excited and optimistic about the future of this field and the transformative impact of our work. If you’re excited too, look at our job postings – we are always seeking machine learning and signal processing enthusiasts.

Stay tuned for future installments in our “What We Do” series, where we’ll dive deeper into specific projects, technologies, and achievements at EXP. We’re excited to share our journey with you and showcase the incredible work being done by our team.

Latest Post